IFR_Net_train.m

%% The is a training Code based on the method described in the following paper: 
% IFR-Net: Iterative Feature Refinement Network for Compressed Sensing MRI 
% Author: Yiling Liu, Qiegen Liu, Minghui Zhang, Qingxin Yang, Shanshan Wang and Dong Liang 
% Date : 10/2019 
% Version : 4.0 
% The code and the algorithm are for non-comercial use only. 
% Copyright 2019, Department of Electronic Information Engineering, Nanchang University. 
% 
% Input:
% imdb: Set of undersampled k-space and the corresponding undersampling pattern as well as label.
%
% Optional parameters in initialize_IFR_NET:
% Rho: The penalty parameters of layer X. default: (1e-3) * 20
% st: The stepsize of gradient descend. default: 0.5
% w1: The weight of convolutional layer C1.
% b1: The bias of convolutional layer C1. 
% w2: The weight of convolutional layer C2.
% b2: The bias of convolutional layer C2. 
% linew: The initiated shrinkage function for layer H. 
% c: The parameter of feature map. default: 0.008
%
% Outputs:
% net: The trained weight.
clear all;close all;clc;
%% add path
addpath('./layersfunction')
addpath('./mask')
addpath('./matconvnet')
addpath('./Test_data')
addpath('./Train_data')
addpath('./Train_output')
addpath('./util')
addpath(genpath(pwd))
config; % set the structure for network: filter number, filter size, stage number, inner block number and iterative number
      ... and the training settings: learning rate, epoch, momentum, and so on
vl_compilenn; % prepare for MatConvNet
%% Load character dataset
load('IFR-Net-radial30-knee.mat'); % load the training data
%% Network initialization
net = initialize_IFR_NET();  % initialize the parameters of network 
%% Train and evaluate the IFR-NET
tic;
net = IFR_train(net, imdb, trainOpts) ; % main training procedure
time = toc;
time = time/3600;
fprintf('The training time is %2.1f hours.\n', time);